I don't know what GP did, but you don't need to write a complete parser. You can write a subset that can find the patterns you're interested in, which is much easier.